Research On Task Offloading Algorithms In Mobile Edge Computing Based On Deep Reinforcement Learning

Mobile Edge Computing(MEC)is a new computing framework that extends communication,computation and storage resources to the network edge,which greatly meets the growing demand for low-latency and computation-intensive tasks of mobile terminals/Io T devices and can effectively relieve the computational burden of mobile terminals/Io T devices and improve user experience.Task offloading mechanism is the key to improve the performance of MEC systems and user experience,which is the focus of current domestic and international research,however,offloading methods based on traditional optimization theory are difficult to achieve optimal offloading decisions within a limited number of iterations.In addition,existing studies focus more on centralized task offloading strategies,while less research has been conducted on distributed offloading methods in multi-MEC server systems.To address the above issues,this paper implements distributed offloading methods under different constraints using deep reinforcement learning,as follows:(1)For the optimization problem of minimizing system energy consumption under task processing delay constraint,a distributed task offloading method based on Double DQN is proposed in this paper.First,a task offloading model with the optimization goal of minimizing system energy consumption is constructed in an offloading scenario with multiple edge nodes,and then,defining task size,waiting time,queue length and edge load as states,offloading decisions as action spaces,and negative values of system energy consumption as reward signals,a distributed task offloading is designed using Double DQN and long and short-term memory network.Extensive simulation results show that the proposed algorithm has good convergence and can effectively reduce the task offloading energy consumption,and the proposed algorithm reduces the energy consumption by 39%and 20% compared with existing methods when the number of mobile devices is 50 and 130,respectively.(2)For the optimization problem of minimizing task processing delay under energy consumption constraint,a distributed offloading method based on deep reinforcement learning is proposed in this paper.First,the latency cost of processing tasks is analyzed in a resource-constrained multi-edge node scenario,and an optimization problem with the objective of minimizing system latency is constructed.Then,setting the system latency negatively correlated with the reward function,a deep reinforcement learning algorithm PERDDQN combined with prioritized experience replay is designed.Simulation experiments show that the proposed algorithm has good convergence and superior quality of service,and the latency of the proposed algorithm to process the task is reduced by 14% and16% compared to the DDQN method for the number of mobile devices of 50 and 130,respectively,and the task discard rate is always below 30%,while the DDQN method has reached 40% discard rate at 90 devices.